activeCycle 2026.1
Systematic Jailbreak Taxonomy and Automated Evaluation for Small LLMs
Benchmarking 7B parameter local models against OWASP Top 10 LLM prompt injection and system override vectors with automated boundary test harness.
POD 05
Secure AI & Model Ethics
Auditing adversarial threat vectors in open-weight language models and measuring demographic parity in automated evaluation pipelines. We build standardized red-teaming test suites mapped to OWASP and MITRE ATLAS security frameworks to detect jailbreaks, indirect prompt injection, and bias transfer.
Core Research Questions
- Which prompt injection attack vectors reliably bypass system prompt boundaries across quantized 7B-parameter models?
- How can disparate impact metrics be baked into continuous integration pipelines for tabular screening heuristics?
- What verifiable watermarking schemes survive temperature sampling and paraphrase attacks?
Scientific Methodology
Automated adversarial prompt fuzzing and boundary probing
Demographic parity and equalized odds statistical auditing
Model card and dataset lineage validation protocols
MITRE ATLAS threat matrix mapping
proposedCycle 2026.2
Algorithmic Parity Audit of Student Merit Screening Baselines
Applying AIF360 disparate impact metrics to inspect gender and demographic balance in automated applicant evaluation heuristics.